Go back

How Enterprise AI Really Gets Deployed

from The a16z Show

1h 20m 36s

How Enterprise AI Really Gets Deployed

Decagon transitioned its AI stack to open-source models to achieve faster, more cost-efficient performance on specific business tasks, such as customer support and sales qualification, where smaller, fine-tuned models outperform large general-purpose models. Despite early use of frontier models, the company now relies on open-source models for 90% of workflows due to superior latency and task-specific accuracy. This shift was enabled by a dedicated research team that builds custom evaluation frameworks to measure real-world business outcomes, not just model performance. Decagon’s long-term advantage lies in its deep integration with enterprise operations—managing compliance, legacy systems, and business logic—making AI deployment secure, scalable, and controllable. The company’s product-led, sales-driven approach allows rapid iteration and transparency, which resonates with large enterprises that fear AI black boxes. As AI evolves, Decagon maintains its relevance by focusing on operational workflows and business process automation, not just customer service. While frontier models remain valuable for exploratory tasks, the company’s model is built to be flexible, scalable, and deeply embedded in real-world enterprise processes. Ultimately, Decagon’s future success hinges on its ability to productize AI workflows and serve as a trusted, transparent partner in enterprise digital transformation.

Transcription

15753 Words, 84865 Characters

English
An AI agent should just be the front door of your business and every interaction whether it's like reactive or proactive with a customer should be handled by AI. This narrative dominated the first half of 2026, which is that anthropic open AI. They're the last startups. They're going to take over everything. Even once you have AGI, agents are going to need somewhere to store work and pull information from and reason about things. I don't think software is a whole in any meaningful ways going away. Unfortunately, the Frontier Labs, they do have small models but you can't really control them in the way that you want. So today, 90% of a workflow is on open source. On the specific task we want them to do, they actually outperform the large, smart state of the optimal. The thing that we built was not an agent that does customer support well, but rather an agent that follows business costs as well. Instead of us having to write these AOPs, do I just do all of that? Let's say we hit AGI and the models can do all sorts of things we can't even imagine today. What's Decagon's mode and why does Decagon like 10 years from now still have a right to exist? The biggest breakthroughs in enterprise AI aren't just happening inside foundation models. They're happening in the products built around them. In this episode, Sarah Wang and Kimberly Tan sit down with Decagon co-founders Jesse Zang and Ashwin Srinivas to discuss why they shifted most of their AI stack to open source models. How they think about deploying AI inside large enterprises and why the future of enterprise software will be shaped by AI agents, not just better models. Hey guys, welcome back to the studio. Thanks for having us. Thank you for being here. Before we get into customer support, I actually wanted to widen out a bit. And Jesse, I'm going to actually mention a piece that you wrote recently that went pretty viral because it's right in the middle of the zeitgeist of conversation right now on open source versus closed source models. And then thinking machines, KBK 3, some very interesting open source models came out sort of right after. And there's this really interesting debate going on around what does it mean to own your destiny when it comes to AI, especially in the enterprise. And what does that evolution look like by use case? So actually, since that's a pretty live topic right now, why don't we start there? So I'm going to talk about our journey first, just to make it very concrete for people. So when we started the company, the goal was just to get something working right when you're the goal is to get something working. Of course, you're just going to use the frontier models because you want to get something out there and have actually deliver value. And so we're using open AI and throw up back at that time. They were kind of like one up each other in terms of how the models performed. And then at some point as we got to larger scale, we started working with larger, larger companies and they had millions of customers. And then we launched our voice agent, right? So a big factor became latency. So it wasn't just like came to deliver good responses, you have to deliver them really fast. And the only way to get latency down, but also kind of make our agent operate the way we wanted to is to use smaller models. And when you want to go to smaller models, unfortunately the frontier labs, they do have small models, but you can't really control them the way that you want. And most small models out of the box are not going to be good enough. That's the task that we want them to do. So you have to find two of them, you have to change them. And so that's when we start looking at open source. So this was about year plus ago. And I mean, it worked really well because if you think about it in the agent, right? So in our agents job is to have conversations. So it needs to do a lot of things at once, right? Like the first step I might do is, hmm, what topic is this person talking about or something else it might do is, oh, is this person a bad actor that's coming in and trying to mess things up? There's all these tasks it has to do. Each individual task doesn't need all of the intelligence of a big model. So all the frontier models are obviously very smart, but they can do a bunch of different things. Like they can do math, they can do coding. You just need them to be good at that one task. And so that's why you can use a smaller model. And if you find you need to be really good at that task, it can be just as good or better than the big models. So that was step one for us about a year ago. We were like, okay, let's start using these open source models. If we took the small ones and then that's why we have now a research team. And it's a very expensive team. We have it because we need people that are really good at taking these open source models and tuning them. So today 90% of a workflow is on open source. And again, the main reason was for latency to really optimize our voice agents. And I think we've just over the last year, we've seen tremendous improvement in how it sounds, how it feels, but still also keeping the accuracy high. And then the remaining 10% of course, we're still using the closed source models and the frontier models for a lot of new projects or new products. And I think that's just where the industry is moving to. So if you kind of were to generalize this every model, you can kind of evaluate along three dimensions. It costs intelligence and latency. And depending on what you need, you want to kind of be at the limit of those three. And sometimes you can trade off. So in our case, we knew that we actually pull back on intelligence, because we all I had to do was that one task. But now we get these latency advantages. I want to push a tiny bit on that point actually, because oftentimes when you see these debates being held on Twitter, the trade off tends to be, oh, do we want the smartest model that is very expensive or can we dumb it down a little bit and get it cheaper. I actually think that is a false trade off, right, because what we've seen in practice is even if you have a quote, dumb or model, you can get it and we've seen this in practice, you can get it to higher performance on that specific task. So when we find to smaller, dumb or models, it's that they're just not as general purpose, but on the specific task we want them to do, they actually outperform the large smart state of the art. So we end up getting all three things. It is better at the task. It is cheaper and it is faster. And so do you feel like today, like you need the most friends here models for really anything at Decagon because your performance is very good already. So we do and we often end up meeting them for auxiliary tasks, right, where when you sort of have auxiliary models to our sort of primary conversational flow, that you have an agent and it's helping a customer with their rebooking or it's helping them. With a process in healthcare, then these are like well defined parts. So we have smart, fast models to do that, but we've, for instance, recently launched do at autopilot, right, which is our agent that improves the core conversational agent. Now for something like autopilot, it is doing a very complicated job, right, it's saying I'm going to go and review a million conversations that just happen. I'm going to try and find trends. I'm going to create variants of the primary model and see which of those variants does better. So now this is a much more broad, open-ended exploratory task. So we think four jobs like that frontier models that are very smart that can try out a lot of things make a lot of sense. Do you think that, I mean, you guys obviously, and you referenced it, Jesse, that you have a research team, right, you guys launched Decagon Labs, but even before the formal launch that's always been a part of your culture. Do you think that enterprises will get there as well on post-training open source models? Oh, okay. What's the timeline? Yeah, I think they'll get there, but it'll probably take longer than people think, because even with our team, finding these models is non-trivial. It's not just, oh, you can say, all right, we made the decisions, use open source, let's just use open source. You have to get the data, and then more importantly, you have to have good e-vals, and if you think about our e-vals, our e-vals are very specific to us. You can't just use some public e-vals set, and like that just does the job. It's like we're testing on our task, and so we have to generate our own benchmarks and e-vals. But I think the point is that at a certain point, it's strictly better to use open source models, because when your use case is solidified and your in production at scale, and you're pretty sure this is the sort of shape of the agent, then there's no reason not to use open source, because you get these latency benefits, and at the same time, you get the cost benefits. Again, we didn't do these for cost benefits, but that's a nice side effect. And once you're there, it's like why use fun cheer for that. But for everything that's new and sort of experimental, or your slash from saying what, or like kind of these products where you really need the intelligence, you're still going to use fun cheer models. And it's just so much easier to use fun cheer models than we're able to employ, use the APIs, and I think that's why in enterprises right now, even though there's a lot of hype for open source, those sort of share of open source inference is actually going down right now. Because people are spending all these new use cases. And if you're spending up new use cases, of course, you're going to use the front cheer models until they're working. But of those use cases, some might die off, but like some might like the enterprises are like, okay, great, we want to keep shipping this in a little bit out. Once it's at that point, they're heavily incentivized to use open source. It's way cheaper and faster. At that point, maybe they can do it in house, or maybe they'll need help from people to help them fine tune it, but that'll eventually happen. I just think it'll be cut us low even right now in our experience. Like enterprises have a lot of desire to move, but they can only do so many use cases at once. There is inertia there and you have to go through all the model risk governance and all the security things. And so I think it'll take time, but it will get there. Yeah, the other reason I think it makes a lot of sense for enterprises to kind of build their cool bundle of labs is that the shape of these models is changing constantly. We don't just build our set of open source models and then done we can move on to our next thing and maybe we'll revisit this in two years. You often need to train new models all the time because as the frontier changes, as the capability of the models changes, you come up with new use cases for them. You find new places where you're like, oh, this toss seems to be getting repeated a lot because now I have this totally new frontier model open source model that now has this capability that they didn't have before. So we find ourselves constantly trading new models and deprecating old ones that are no longer relevant because maybe the front. Venture has advanced a lot, the open source Venture has advanced a lot, and the model out of the box can do a lot of things that it couldn't do before. So because the model landscape is changing so quickly, Daccanon labs is in a way a model factory, a source, right? We've really built it to compress the time between new model coming out and useful fine-tuned to our task model kind of popping out the other end, just because it happens all the time. What do you guys think about what to in-house from a talent perspective versus, there's also a pretty broad ecosystem right now that is, you know, could be RLs of service, e-vells, et cetera. Like, what's your framework for, hey, this is mission critical and we need to do this best versus, yes, it'd be great to, you know, outsource this. You know, in practice, we've seen that so many things, relevant to model training are so tightly coupled to the use case that we have, right, that we find that we end up needing to build a lot of tooling internally, but when we have open source models that we want to fine-tune, we find that if we can clearly tailor our e-vows to customer outcomes, it's way better than just looking at like loss curves over time, right? We're not just saying, oh, can I do this one specific task? We're measuring the entire system end to end. We're not just saying, is this model good at this task? We're saying, is this model working in concert with all these other models delivering the end customer outcome that we care about? And because that is so unique to our setup, we've found that in practice, we've needed to build a lot of, a lot of the infrastructure that we need to train these models and evaluate them. Now, for other things like getting labeled data and measuring the diversity of our data sets, we're like, yep, these are tasks that are common across possible companies, in this case, we want to buy things from other vendors because that'll just help us get those models to production posture. Ultimately, the only thing that we care about is how can we get the best model to production as quickly as we can? And so it sounds like, you know, there's a lot of talk right now about like tokenomics and how expensive it is to actually run a lot of these models. Based on this conversation, it doesn't sound like you guys spend that much time actually thinking about the costs of these models. Is that correct? It's really about the performance for you. This latency and accuracy is definitely the driving factor for most of this, right? Cost is a nice benefit in that, you know, surprisingly, this is one of the few like tasks where you kind of get all the things for free, right? Like, we don't actually have to trade off costs and latency and performance. And so just by optimizing for the drivers actually latency and performance and we just get cost is a nice side benefit. And do you think that's like something that's unique to the way Decagon is run or do you think there's something about like this conversation about tokenomics? That for some reason just doesn't affect you guys in the same way. No, I think if you're a growth stage company, then you're honestly like obviously we have to be responsible with our costs, but that's not the highest priority, right? The highest priority is just growing. And when you're talking to a customer, they don't really care what your cost are. They just care about how well your agent performs. So that's, that's the main thing. So yeah, actually, if you look at like our unit of output of our agent, which is in our case, a conversation, that's really what our customers care about. They don't really care about how much that conversation costs to us or how many tokens are in it. And actually over time, the number of tokens we're using per conversation has gone up because we're actually doing more model calls to make the quality better to do more checks to paralyze more things. And that's just the state we're in right now. Like eventually, you know, if we've won the market, then like that's when it's like, yeah, you know, now it's time to look at our cost and see if we can optimize further. But it's just not the type of priority. I also think we as a company are at a different stage than a lot of people that are talking about tokenomics on X, right? Because the tokenomic debate really is around, hey, I'm on a frontier model today. Should I, like, how do I analyze my cost and should I move to open source, right? And there, I actually think it makes complete sense. If we ran our entire business exclusively on frontier models, I would care a lot about cost and I think a lot about that. But once you've already made the jump to saying, okay, now we know how to think about open source models. We know how to decompose upon. We know how to build, train, and deploy these models very, very quickly. All of a sudden that the cost aspect becomes a lot less pressing. But if you're still in the world where you're using frontier models for everything, then I'm using thinking about two of us makes more sense. I just want to make this meta observation that a lot of the conversation, even just now, has been about things like training models, right? We're talking about reinforcement learning. And it really does blow away what I think is, you know, a lingering misperception about what an AI application is. And just to get into that debate a little bit, because I think it's sort of this narrative that dominated the first half of 2026, which is that anthropic open AI. Applications. They're thin UIs with FTEs, you know, with implementation attached to it. You know, we talked a little bit about deck gun labs, but can you guys just share what is your, like how do you think about this potentially false dichotomy of app versus infrastructure company? Clearly you guys are so much more than the UI or the implementation. And, you know, does agent lab, which, you know, popular description floating on the last couple of weeks, like, does that describe it? Like, how do you guys think of deck gun? So, I'll give a quick perspective, just like from the, from the POV of like a enterprise, and then maybe we can talk about like broader industry. So, let's say I'm like a Fortune 100 company, right, and I'm looking out there on all my use cases, and I have a choice of partnering with an application company or sort of using the labs and building from scratch. I think there is a lot of merit to partnering with the labs in certain cases. I think if you look at our case, right, like we just talked about all this fine-tuning stuff. I think a common misconception that people have is, you know, fine-tuning is a way to like customize it for that customer. In fact, most of the fine-tuning we do is like customizing it for our use case, like the customer service use case. Now, and it's worth it for us to do it, because that's all we do, right? We do these agents across all of these different customers, and so it is worth it for us to put in a ton of time and research into like, how do you tune this one model to be good at selecting customer service topics. But if you're the enterprise, is it really worth your valuable research resources to like tune a model for these customer service behaviors? Probably not. So, that's like, that's one reason why people partner with applications. Another reason is, let's say I do put in the engineering effort to build like a, some agent myself using the frontier models, and, you know, to my earlier point, you know, I'm not fine-tuning for behavior, but I'm sort of teaching the AI my own procedures. And again, that doesn't happen through fine-tuning, that happens like in context, because if you were to fine-tune on that, you'd have to reverse it every single time, you change your procedures, which doesn't make sense. And so you kind of build out your logic and you're building your business logic in, well, then you launch the agent, and then the second day you look at your conversations and you're like, oh, well, actually, I need to change these three things, and now it's more engineering effort to do that. And it's like constantly engineering effort. So I think people will partner with applications when the use case calls for a broader platform, where there's a lot of value in using, you know, the stuff that we've fine-tuned and using the software stack we built on top of the models to capture business logic. And that stuff has nothing to do with the models, right, like how, how the business logic gets captured by this AI, like how do you handle someone calling in because, you know, they're flight was canceled and they need to re-buck three people at once. It's like, that is business logic that the AI needs to know and you're encoding that, but that has nothing to do with the models themselves. And so that has to exist in the application there. I think that's where applications will still shine, because you still need the application there. And it's not so much the models. And the labs themselves will have more application capabilities, but those will be fairly general. Like they're, you can maybe build general agents that can do this thing or that thing. But for a lot of these like core verticals, like ours, our, our thesis is that, you know, you're going to need something that's like very deep and has all the integrations, has all the ability to capture business logic, has the ability to run, you know, tests and experiments and then review the conversations and run QA and like, you know, have tooling free compliance team to monitor like what's happening. So that, that's our thesis on it. And so it's kind of, it's, it's not black and white. Like there will be some use cases where it does make sense to use the, the frontier models, but there will be these like core verticals where going super deep makes sense. Yeah. I also think, you know, everyone is kind of bleeding into everybody else's space. Sure. Right. Lots of convergence. All the, all the labs are building applications on top of it because rightly so they're saying this is how the enterprises dealt up some more, right? And this is how the enterprises see ROI from using our products us on the application layer. We're realizing that we can squeeze out a lot more performance and latency and cost for the use cases that we care about by building our own models, right? Which is, and I think this, this split, this kind of bleed over makes sense. And I think it'll continue. I'm not as bought into the, the labs are less startup view of the world though. And I think this is true both for SaaS companies and for the new AI startups. Because in a way, we human beings are kind of AGI, right? And human beings have needed to use software for lots of things, you know, you need databases to put stuff in. for the United States. RMS to truck things, and I think even once you have AGI, or AGI agents are going to need somewhere to store work and pull information from and reason about things. I think, you know, a certain plot, subsass companies that were solely built for people to do work might face a bit of heat, but I don't think software is whole in any meaningful ways going away. And I still think, you know, on the application layer, there's so many different kinds of work that need to be done, that can be done faster, more efficiently, more cheaply. So I think there will always be a space for application layer companies. Maybe in the long term, application layer companies just become labs for specific verticals, you know, because your primary product ends up being the models that are just really good at doing those specific tasks, but I think the application layer probably here to stay. I also want to kind of pick on another thing that you kind of said in passing. Yeah, please. And maybe this is like a spicier take of, oh, our application layer companies just forward deployed companies that are kind of doing the last mile of work. Yes, right. Well, I was saying that's a misperception, but I think it is a common one. Yeah. I think it's a really hot thing, also, again, not to pick on tech Twitter to be like, oh, like, you know, we need to bring back the forward deployed engineer and, you know, every company is hiring tons of forward deployed engineers. I think this is a trap. Oh, same work, okay. My view on this is that for deployed engineers or necessary or newly necessary for early stage AI companies, because the workflows are new, right? If you're building a SaaS company five years ago, most SaaS products are pretty well explored, right? Like, you roughly know what the user is trying to do, and your job is maybe come with a slightly cleaner workflow, but broadly, you know what the users are trying to do with a design app or a CRM or something like that, because those workflows have been explored with AI products. Nobody knows what the workflows are, because nobody's used these things before. So a forward deployed engineer in this case is honestly just embedding with the customer to learn the workflow for the first time as the customer learns the workflow for the first time. And they're kind of paving the road, like, you know, they're kind of laying out the track as they see which way the train is going in a way. But long term, I think they should just be building a product, right? Like once you know what the workflow is, you should not be relying on forward deployed engineers anymore, because once you know what the workflow is, if you can productize it, you should productize it and then become, you know, a typical company with these scaling properties of a tech company, and if you can't do that, then you're just building a glorified consulting truck. Well, I looked at dig into this more, so I know when we started working together, probably almost exactly three years ago, and you guys landed on this idea, there were two things that were relatively contrary and that now feel kind of standard. The first was when you started an AI customer service, a lot of people were like, that's a GPT wrapper, and we talked a lot about like, why that's not the case. And the second thing you did was say like, hey, we actually want to do the work ourselves. Like, we don't want to just be a software platform, and now we have the term for that that's like, AI agents and everything, and you've popularized agent PMs and forward deployed as a new type of role or more common type of role in Silicon Valley. But you're like agent PM slash forward deployed team has actually evolved a lot since you first got started to now. And so we'd love to talk about that. Like in the early days, what did it actually look like? And then as you started to learn about these workflows and we're able to productize them better, how is that function actually evolved for you guys? Yeah, if you look at a lot of our forward deployed teams, they are building product in one way or another, right? So all of our forward deployed engineers, for instance, build core product, right? Their job is to go in and understand as we're working with an enterprise, what are the things that they need that the product does not do today? But the output of that is not a one off thing that is just built for that customer. It is something that is contributed to core product in a way that the next 10 customers that ask about the same thing get it for free, right? Similarly, our agent PMs are working with our customers to understand, you know, how is the product broken today? How can this actually be deployable within an enterprise? What are the new things that we need to build to our core product to make it deployable within an enterprise? So at the end of the day, all of this boils down into product improvements, either through actual product improvements or through honestly process improvements, right? Like we work with very, very large enterprises and we help them through the journey to go from, okay, this is what your work looks like today. Here is how we can take you through changing your processes, through implementing new technology into this world where agents are doing a lot of work for you. So a lot of their product work is also processing a lot of how we make that, how we help accompany through that transition. And also, when you used to be a deployment strategist at Palantir, so you're very familiar with the forward deployed model that Palantir popularized. How different is that at what you did at Palantir versus the way you conceptualize this role of decadentance? I do think people use the term FDE very loosely and so on and out of that way. Yeah, and I think it's dangerous to mix the two of free consulting work versus actually doing product, you know, Sham, who's the seat to Palantir today, had a phrase, "Internally, I think now it's been written about a ton." He would say, "For deployed engineers, eat pain and extract product." Yeah, I think it's a massive misconception because people are like, "Oh, if Palantir is such a hot company and they're doing so well, this is like a cool take on how to implement stuff." First of all, very few companies, if any, can do what Palantir does, which is like closed massive deals off the bat and like it's kind of worth it to spend all that effort. So I think a lot of the people that are doing this super forward deployed strategy in like, "Hey, we'll do any AI use case for you." They will eventually have to reckon with like, "Okay, can we find a product that scalable?" Otherwise you are just building like a modern, like, accenture or something, which yeah, it could be good if that's what you want to do, but I think people kind of conflate the two. So hey, I'm building a hot, like, software company and you know, we also have FTEs and like, that's our big strategy. So that's something that we've generally been very mindful about from the beginning, is that we really view ourselves in our core. It's a product-led company, right? In the sense of, you know, we're building a core product that everyone can use and we're not just here to, even though we have a large team now, just like build whatever use case that pops up because, you know, ultimately everything has to go into the core product, otherwise you can't scale. At the same time, of course, you know, I should say we also kind of sales led in the sense of like, the product is informed by sales. So we're not sitting there and just, you know, coming up with product to build. And so you kind of have this system where, as Ash was saying, we have all these people that are, you know, forward deployed in the sense that they work very closely with the customers, their job isn't just the spin up like random new use cases here and there and like just doing whatever the customer wants because in the short term, that could actually eat so like larger contracts and you can kind of hunt for wherever the pain is, but that long term is just very difficult to scale. So their job is to kind of compile all those learnings from sales into a core product. And the goal at the end of the day is like we should have the best product out there and like be able to iterate on that faster than anyone else. And that's in our space at least our vision of like, hey, the winner in this space is going to be very product-driven. I actually want to pull on this thread of, you know, that you talked about where your FDs are actually approving the product, you know, Decon is a very product-driven company. So just to kick off with a very small, seemingly unrelated anecdote, but I was just trying to convince someone on the broader team to not leave A16Z for a Frontier Lab. And one of my arguments was that there was a good long-term career here. And the response that this question gave me was, we'll have AGI, we don't need careers in the long term. It really hit me because I thought I was AGI-pilled, but I really not thought about from that perspective. And I'm curious, you know, we talk about the product improving, right? Can you make that concrete for us? Like, what are some of the OSHIT moments as you improve your product for your customer, either from your end or your customer's end, if you can share on like, oh my god, I didn't realize AGI could do that. And then, of course, I'm going to ask you your thoughts on AGI and what that timeline looks like. Because you're in the nitty-gritty trenches of the enterprises using AIs. So you may have a different perspective than, you know, we do. The first thing I want to say is, I'm like, certain there will be careers after AGI. The reason for that is, like most of our jobs, more sure are for jobs, kind of like made up. Most jobs are made up like a very real job. Unless you're like building infrastructure or like growing food or something, like most jobs are kind of like layers of abstraction, guilt on top of like other stuff, right? And that isn't to say like the jobs aren't valuable, it's just the kind of made up. So when AGI is here, like it will change people's jobs, but people will stop jobs because you're still going to do things for other humans and whatever. So I don't really believe that a career will be gone after AGI, I don't think people are just going to be sitting around. I'll say one observation, which is like for me, it was definitely, um, do it. So early days when we were building the product, right, like the core problem we're solving, again, being back to CL's lead, we talked a bunch of customers and they're like, yeah, the value we want to get out of what you're building for us is that, you know, we can put in front of customers, they can have conversations, and it's giving them a much better experience, and also it's like way easier for us operationally, right? It's like you're saving cost and you're making customers half year. So that's like the first agent that we built them. And that was like the core agent we worked on for the first year, year, year, two years. And the agent on our end, when we were building it, it was just sitting up a ton of stuff. Like we would have to write these procedures, and we kind of create our own format of procedures, so we call them agent operating procedures that teach the AI how to do things. We have to write these tools that the procedures can use to access systems and pull APIs and whatever. And then after that, we need to like create all these tests to make sure that this thing is working well and that you can kind of simulate all these different situations. And afterwards, once they're in production, like we would manually be reading conversations. There's like a ton of work that goes into it, even though the core product we're building is itself an agent. And so what we do at is, it's kind of a separate agent. It's like a second agent that's much bigger and much slower. But its job is to do all the tasks I just described. So now, instead of us having to write these AOPs and write these integrations and tools into their systems and write these tests and monitor the conversations, do I just do all of that? So it's like a second agent that is smarter to do all of these things. And it just feels very magical, 'cause you can just literally tell it like, hey, I have nothing built yet so far, but here's a bunch of transcripts I have. And here's some documentation. Like you go figure out the best way to do these, all these procedures I want. And it'll go do it. And then of its own accord, it'll also write the tests and simulations that go along with those. And once you're done with that, and you actually put in a front of customers, it'll be the one that's monitoring all the conversations and it'll flag things where things are going well or poorly. And it'll say, yeah, I read these 1,000 conversations and actually there's this one topic that we do really poorly on. And I've noticed that. And I've also drafted these improvements for you. So it's like, wow. It's like one agent that can do all of that. And so that's very magical, because well, first of all, it was not possible when we first started the company. It only became possible when all the reasoning models got better. And of course, anthropic open AR, making these reasoning models mostly for like the clawed codes of the world. But they are also really good for, like do it, for example. And that was kind of like a moment where it's like, oh, wow, like first of all, you can just see the improvement over time of the models. And two, it can do all these tasks where it just would not be able, you would not expect an AI to be able to do all these tasks at once, but it can, it can do them very well. And I think that's, that was like a very visceral moment of like, oh, wow, the models are getting a lot better and they're becoming like very generalized. Like they can do all these things. Like clearly, the models were not trained on like our specific task, which is writing these procedures and writing these tests, but they're still good at it. - And also, to your other question of how, you know, how did we come up with all this and how did the product improve? Every single thing that Jesse just talked about was the result of four deployed people doing things and I was figuring out how to productize it, right? So for instance, we realized that, hey, when we go into a new customer, we need to spend all this time writing up the AI piece manually. And we're like, wow, this is quite a lot of time. How do we productize this? And we built that into duet. And then the second part, which we call duet autopilot, was once we built duet, you know, people were using duet to write things up and then we're like, oh, well, there's still a lot of time that goes into iterating upon the agent, right? Once it grows live, like reviewing conversations, figuring out how to improve it. And we're like, great, let's productize that as duet autopilot. In fact, AOPs themselves were a result of this exact scenario because before AOPs, you would have to write all these procedures in code. And then we found that, oh, it's taking a lot of forward deployed engineering work to write all these things in code. Why wouldn't it be so much easier and efficient if we could productize it by writing it in plain text, right? So the way we think about my product improvements, forward deployed engineering investment, everything is built around what can we productize from forward deployed work so that engineers and any kind of customer facing resources on our team don't need to be as heavily evolved. - Can I ask you a blunt question on this line of thinking? So I know we've already talked about why open-and-a-thropic won't be the last startups. And you've talked about why there's room to have much more specific use cases and specific companies. But you've also had like, oh, shit, moments where you're like, these labs are getting so much better and the models are getting so much better. And so long-term, let's say we hit AGI and the models can do all sorts of things we can't even imagine today. What's Decagon's moat at the end of the day after all of that? And like, why does Decagon like 10 years from now still have a right to exist? - I think in the short term, actually, it is the ability to work with enterprise resources. And what I mean by this is that the capability of models today is far greater than they are being used for within the enterprise, right? By which you can't just take a model and say, I'm just going to get this model access to everything within the enterprise. And I'll just figure everything out, right? Like that's practically not how these things work. So to make models like this deployable within the enterprise, let's assume that every model is just perfect and makes no mistakes. But to make something like this deployable within the enterprise, you need to say, okay, I need a way to be able to tell the model what it can and cannot do and make sure that it cannot do anything like catastrophically wrong. Then I need a way to make sure that, you know, hundreds of people within the enterprise can collaborate to make sure that the agent is behaving as expected in the use cases in which there are experts. Then I need a way to be able to test this model and make sure that it doesn't cross any like regulatory lines that I have and test it, make sure it works well. Then I need a way to, you know, look over the, you know, millions of conversations that happen for me to extract insights for the rest of my teams, right? So there's a lot of just infrastructure and software that you need to build around these models to make them deployable within an enterprise, to make them work with all of the legacy systems that these companies have. And so I think for the next few years, that's probably going to be, you know, the primary thing that these models need to be able to work. Now, once that gets commoditized because the agents can build that on the fly, that I don't know and will create in three years from now. - By the way, I think just listening to you guys, it's very clear that you guys deeply understand how to sell AI to the enterprise. And I don't just mean, you know, mid-market, newly IPO companies, I'm talking about some of the largest companies in the world. And I think this is particularly interesting because you both are very, very technical, but also go to market animals. So I want to talk a little bit about that. But you know, it just feels like you've turned what maybe started as feeling more like a David and Goliath with multiple Goliaths market dynamic to really a two-horse race between you and Sierra. So share a little bit more about why some of the largest enterprises in the world are buying from you guys. And I think what's really remarkable to us is people have studied application software for, you know, over a decade. The sales cycles are crazy fast. I mean, I'm sure you guys with urgency want them to be faster, but like, usually you don't sell contracts that big to enterprises that big, right? That takes two years sometimes. So maybe just share a little bit more on, there's a long question, but like, how did you grow that commercial? Did you come with that into founding the business and what's resonating with these large companies that enables you to get in and move self-quickly? Yeah, I mean, we have a lot of respect for Sierra and also just, you know, other folks in the space, generally the big platforms, like they all, the platforms themselves move a bit slower, but I think that we've, we've met a lot of those teams, they're very competent teams, they're optimizing for a lot of different things at once. Design conference of Sierra, I mean, our most recent customer actually, Trin author Sierra to come to deck it on. And sort of the reasoning when we asked them was, it kind of goes back to this like deployment model. It's like, you know, when they worked with Sierra, it was mostly FDs and it just felt like a black box where, you know, the FDs were good, but they had to go through the FDs for everything, so to build new journeys or to, even get like a deeper understanding of what was happening in the conversations. And then over time, that kind of just created a lot drag on how quickly they could move, right? So in the initial deployment, that was good, but then over time, maybe FDs were staffed to other things and it just took them a while to get insight into what was happening and then build out new journeys, right? So over the course of the year, they may be built out, I think they said three. And so the reason they came to us is 'cause they, they had that frustration and they wanted to kind of have a different model where it was a lot more productized, right? Back to us being like very product driven in our vision, it's, they should have a core product that, even if we're there helping them, even if we are for deployed, it is like a, and in service of helping them build a product that they can use themselves and iterate really fast and kind of have everything in their control, right? So we like to call this like a glass box approach instead of a black box. And yeah, so within basically a month, they spawned up like seven new journeys on decking. - Wow. - Yeah, I'm just, - And it took in three, a year to get three. - Yeah. - Okay. - And so it's just kind of the speed of iteration. Like some teams will really like this. Like hey, we have control of it, we, our teams, especially our non-technical people can come in and do things and they understand what's happening and in the conversations and maybe there's other teams out there that actually do like the like, hey, you guys do everything for us. And, that approach. But that's kind of the difference in their approaches. And that's why we've been, I would say, having a lot of success there. And then zooming out, just selling to their prize. Yeah, I mean, this is something that, you know, Asher and I have never sold to their prize before. And I think it just happens that both of us find sales exciting. And the enterprise, it was kind of a quick learning curve for us. So I don't think, I think a lot of it, honestly came kind of naturally just because our space is so hot. And like generally in these conversations, we're not really having to convince people to like invest in this space. It's like more of a, we're the right approach for you. So like you partner with us. And yeah, with the enterprises, it's really just about like navigating the orgs and really having empathy for what they value and what they're afraid of. So yeah, I mean, we just back to being sales led, right? Like we always from the beginning were like, hey, we're going to be like extremely strong on the go to market side. And that's going to inform the product. Even though we have this product driven philosophy, like we don't want to just be dreaming up random products to build. So we always had that DNA. And then in the early days, we were just, yeah, pushing really hard. And I think, I also want to say, I think we got kind of blessed with a really strong early sales team. And we have a wide, really talented people in that group. And that helped us really get leverage as we were talking to, you know, the big enterprise. Some of whom cold applied to you guys in the early days, I remember. Yes. Yeah. Yeah. Yeah. Cold, cold applied. Because I think they were working in the in the space already. And they were seeing from a far, like what's echo gone was starting to do. Yeah. Yeah. So some of them have like non-traditional sales backgrounds, you know? So they had non-traditional sales backgrounds that come into sales. The other profile we had a lot up in the early days was like, like I believe athletes, I guess. And those profiles were kind of a good foundation for the group. And you know, we've had the scale that team really fast, which is never easy. So there's our things that we're still trying to catch up on in terms of enablements and in org structure. But because we've always had that intensity on the sales side. And the whole company knows that, like, you know, everything starts with sales and it kind of propagates back. And we've always had that focus. The, you know, the other thing I think that helped us a lot, you know, to your earlier point about how did you get some of these large deals close so quickly was, I think we were very curious about how we could productize parts of it within the enterprise. Yeah. Right. By which I mean, we aren't a company that just says, hey, here's a product. We'll throw it over the wall. And, you know, you get it a year later. Because within a lot of these enterprises, the question they have internally in addition to will this product work for me is, can I actually get this live, right? And with a lot of these enterprises, it's actually complicated, especially if you're in financial services and you're regulated, for instance. So we actually spent a lot of time sort of mapping out that part of the journey very well. So that when we walk in to one of these enterprises, we can walk them through in very granular detail, how we go from this first meeting today to going live at 100%. So, yeah, you know, at this stage for a company like you, this is what your model process is likely to be. This is how your testing process should look like. This is how we should do the initial roll-on. This is how we should catch any issues that come up and how we're going to fix them and how we'll ensure that they don't happen again. So the product and technology part of what we sell is important. But for these large companies, equally important is us helping them think through the process to actually get this deployed and at scale. Because I think that is something that often gets overlooked by tech companies selling into enterprise. Yeah, for sure. And by the way, I can also personally attest to the strength of your early-go-to-market team having met a bunch of them. That being said, I don't want to underplay how many times a decision-maker has told me that one of the reasons of many that they're going with Decagon is they want to make a bet on you guys. They'll literally say we think the founding team of Decagon is going to move the fastest. This is a very fast-paced market. It's changing on a weekly, if not daily basis. And we think that you guys are going to look at the chest pieces and make the right moves because it's not sort of like a, oh, everything's going to be the same in a year or even six months type of market. So I'm curious. I mean, don't give up too much alpha here, but how much time do you guys spend on sales as founders? And how has that evolved over time? I probably spend most of my time, like 80 percent maybe. Yeah, okay. And yeah, I think a lot of it is just pushing speed. So some of it is like, hey, like as one of the founders, you just have to be in calls and like, you know, people want to meet the founders, but also is just how can I be the sort of the main force that's pushing our team to go faster, but also just like the partnership to go faster. One of the things with that is like, you know, now we work with, you know, several of the largest, you know, banks in the world and airlines and and telcos and no matter how fast you go, like there's all there's, those are still like massive organizations and it will take time. And so one of the things that we really try to do is, you know, kind of take the project and piecemeal it. So we're not just deploying across every surface area, every use case at once. That's really just pick, you know, one or two of the top use cases and just get a win there. And I would say that's been helpful, right? If you navigate like a big bank, like it's, it's not going to be like a quick, like, you know, just to close a big deal. At least for us, it's not me. Some people can do it, but it's, yeah, it's still a lot of effort. It's, it takes time. And so you have to figure out ways to, you know, design things from a process perspective and an oral perspective and in our product perspective as well, that like just keep shortening that and like finding clever tactics. And that kind of takes founder involvement. You want to really expect a, like a sales team to just like, constantly be coming up with new configurations there because, you know, they're not responsible for the product, they're not responsible for, you know, the end-to-end, you know, process that the company runs. So having the founder involved there is very important, I would say. And then the sales people, their job is kind of execute on the sales side, like build champions and navigate the org and so on. The other reason to also spend a ton of time with both sales products and existing customers is also because the market is changing so quickly, being able to really quickly understand what are the things that we are not doing today that should be very quickly, right? Because model capabilities are changing all the time as models roll out people seeing other things that happen. They're like, oh, wow, this was really cool. And like I'm seeing this here and you guys aren't doing that as well. So being able to like stay really, really close to that feedback loop because that's something we never want to, never want to let go of. Yeah, don't talk about how that like informs product roadmap, meaning probably like two years ago or so, most of Decagon's customers were focused purely on customer support. And nowadays, I would say that's actually not true. You guys have broadened your vision to be like an AI concierge for your customers. Like, don't talk about what the distinction between those two actually is and in practice, like what that means for both your sales teams and your product teams. So, you know, when we launched the regional set of use cases that we sold were in customer support. And the reason was because one that was one of the biggest challenges that a lot of our early customers were facing. And two, that was where the capabilities of the models ended at the time, right? Like that is all, that is the pretty much the limit of what they were capable of doing. Now, however, as models have gotten better and our customers have realized, well, why would I have one set of models that just learns about my customers when they have another when they have a problem and something else when they come to me to buy something, right? And so, we had a customer that we were really when live with them for customer support. And then they realized they were like, well, you know a lot about our product now. You know about the capabilities that it has because you need to do that for customer support. You know how we like talking to our customers and our brand, um, can help us with inbound sales, right? When someone comes in, answer questions about us, do some discovery, and then, you know, assign it to the right enterprise rep if it's, you know, a deal of large enough value. We had another customer that started using us for a lot of operational workflows, right? So we are able to now proactively reach out to them once we start seeing any kind of issues on that on that customer's account. Because ultimately, at the end of the day, the thing that we built and we kind of built this intentionally from the start was not an agent that does customer support well, but rather an agent that follows business process well, right? And executing on operational workflows, doing sales lead qualifications on surrounding customer support questions at the end of the day is just an agent following a business process. And we kind of built it flexibly enough to kind of do all these things because we realized that hey, at a certain point, the models are going to get better and they have, and what do they specifically get better at that allows you to do that? It is specifically the ability to follow instructions well, right? So when you had, um, models, you know, let's say a few years ago, you'd have to give it very, very tight guidance, very specific instructions that you didn't want it to deviate from. And as the model's got smarter, you could kind of give it broader and broader guidance, bigger and bigger instructions. And just trust that the models have good enough sense to interpret it. like a human word and kind of fill in any missing gaps. Right? Because for, again, for customer support, you can have a very tight pot that a model should pull. And that's all you really need. Whereas for sales qualifications, you kind of want to ask open-ended discovery questions. The conversation is going to kind of bob and weave. And so you need the model to kind of fill in with reasonable things. So that's specifically what the model got better out of the years. Yeah. I think if you think about, you know, what is the 12-month product roadmap? Because things are moving so fast, like the real answer to that is like, you know, kind of like see how it evolves. And we obviously know what we're working on now. But I think realistically in today's AI world is very difficult to have like a 12-month roadmap to a tee. You maybe know how like some themes of what you want to build. But ideally, if you have those things, you just build it like right now. Because it's so, so fast to build things now. So I would say that that is that is one element. But then the sort of long-term vision is still very clear to us now, right? Which is we use the Trump concierge. But really, it just means like, hey, an AI should just be the front door of your business or your brand. And every interaction, whether it's like reactive or proactive with a customer, should be handled by AI. And we've already seen that AI is very good at that. Right? Customer service is like a huge pillar of that. We saw all these inbound interactions. But why not have also be able to do all these other things? So over time, again, we're not trying to figure out on our own what these things are. We kind of have now have a lot of customers that will give a signal on the things that they care about. And those will be the things that we build. We've talked a lot about how the models are getting better. And obviously your capabilities are moving along even ahead of that progress. What are some of the bottlenecks right now that you're seeing? Whether that's on the capability side, I don't know, it could also be around persistent memory. I don't know if you guys feel like that's kind of up to snuff on where you would like it. Or could be other bottlenecks. But curious, what would you like to see and what's kind of holding back? Hiring. But it's less on the AI side. It's actually people. We are we are voracious consumer of tokens, but we've always loved more people. I think there's so much to build these days. Yeah. And why can't you hire AI agents to do the things that you're doing? Yeah, we are voracious consumer of token. Our token bills are very low. But there are still things. I don't quite yet think we're at the point where we can have the AI agents make decisions on what to build and kind of have that. The taste of is this done yet. So we can outsource a lot of specific execution steps. But I don't yet think they're the point where they can make the call on what to build, what to exclude things like that. So the models improving have the sense like you found it three years ago, has it changed your hiring needs at all? Because a lot of people are like, oh, you can build like one person unicorns. Yeah. I think this argument does get tossed around luck. But I think the the an easy counter example to this is all the the AI coding started to hire like crazy. You know, they're like the most sophisticated users presumably than they are hiring like crazy. I think the reason is just because everybody has access to these tools. And so if our competitors are going to use them and build more things, we need to build more things right? If somebody else said, oh, here's our roadmap and now we can get through it in, you know, third of the time. And then they just stop hiring. We would just take that to mean, wow, we can get through it in a third of the time. Great. Let's build three times as much stuff. And turns out everybody, you know, does the same calculus. So everybody both needs to keep hiring more and shifts more. So it is great for consumers of these models. But I don't think it has material change or are hard on plan. It's so funny. I thought you're going to say something like I don't latency of voice models or something like that. But it seems like on the technological bottle next side, it's yeah, there's still stuff that we're waiting for, right? Like, you know, voice to voice models is an interesting frontier that there's still research happening on. You know, the getting smaller models to be smarter smart out of the box, right? So there's there are going to be still developments that we are watching closely that we care about. But from a business perspective, it's less on the model side and more on the, just like, can you build the company fast? Yeah. Actually, maybe since we're on the topic of hiring, and, you know, I feel like grindslop has become this theme on X. And it's so funny because, you know, as a, as a VCE, like I'll admit, you know, when I, I mean, you guys are famously in the office six, seven days a week. I hope this isn't get marked as grindslop now. But, you know, I think constantly hustling for your teams and your customers. And it's so funny to see that kind of turned on its head. And so I'm just curious, what is your, what are your thoughts on the narrative out there? And like, grindslop is such a general term, but like, you know, you guys grind, but there's also a lot of camaraderie and excitement in the office. So like, just, you know, say more about that and like, how you're building the culture. Yeah. I mean, grindslop's a funny term. So we have never posted grindslop because we kind of view that like working hard is just like, I don't, I don't think people are like working hard to grind. It's just like, oh, there's a lot of stuff to do. Like, we kind of, people spend time there, mostly because like, hey, it's, you know, this is like a fun time of our life. Like, we want to take a manage of our talents and sort of potential so that it's not wasted. I think that's, that's the main reason. And it's a very sort of like, it's like many orders from like the main goal, which is like, can you build a good product and you went into space? And, but it's like one of the effects of that is that people work harder. And even from the beginning, like, yes, we have an office culture, but like, we're never mandating people to come out on the weekends. We don't really care how long they are in the office. It's just people are in the office just so that like, we can maximize communication. And that's how we view it. Like, I don't view ourselves as like, you know, abnormally grindy or abnormally ambitious. So I think like, we're, I think, we just, like, I have live ambitious people around me, like, ushering in all these people. So it kind of just becomes normal. And like, everyone works hard. So it's kind of like normal. So I don't think it's like something that we view as, you know, something like super special. Yeah. And it kind of also view a lot of what we do very much as a team sport, right? Where we very, rarely have lines between our orgs in that you will very commonly see engineers on early stage sales calls. You will see sales people like debugging parts of the product. You will see our APM team in absolutely both ends of the spectrum. And so I think being able to have teams that are so kind of disparate from like a function perspective all working together kind of one kind of needs people in the office because everybody's kind of jamming on ideas together. But to it also makes it fun in a way because, you know, everyone is like working together towards some very specific outcome, right? It's either getting this building this thing for this customer in time for the deal to close or launching this new thing. And because it's so many different teams working together, it's kind of a, it's kind of a we're all in this together to ask across the line. Yeah. And how does that scale as, you know, you've launched a lot of new offices recently, you've a new Australia office, London office, your New York office is growing like crazy. Yeah. How do you maintain that culture when you guys are both in San Francisco? I don't know. I mean, I don't think it's a solved problem for us, like we're constantly working on it. And like every single time the company gets to the next days, there's like new things we have to institute to just make sure that everyone understands the culture and there's like very high accountability and, you know, there, there is pressure because there should be. And so that's something that we're constantly adding on because in the first 100 people, it doesn't really matter because everyone kind of knows each other and like everyone's, but like as you grow, people might not have no one's communicated the visions of them or communicated the culture to them. And yeah, I know like Ben was talking to us about the A16Z culture. And it's like how it's very like action oriented. And you can't just like put, put like fluffy stuff on there. And so yeah, it's all these things that we're trying to do a better job of as we grow. The other thing is also we bring everybody that we hire out of San Francisco for a couple of weeks when they start. So they were kind of immersed in the kind of the original kind of soup of culture. And then for every new office, right? At this point, New York and London, all these offices are big enough that they have the original Declan culture there anyway. But for brand new offices, we actually have people from one of these hubs go out and spend a few months there really until the office becomes big enough and has its own kind of culture so that it doesn't kind of become too different from what kind of the original Declan culture was. Actually, on the topic of international work. So we brought on Ragu Raghuram last year, obviously former CEO of VMware and in addition and investing, he's helping us build out a ton of our international capabilities. And I think one of the things that we've seen is that our AI companies are just getting pulled internationally way earlier. Like for you to have an Australia office, it seems premature, except for the fact that you have customer pull, your scale is quite large for, you know, how long ago you were founded. But maybe say more about that. Like does your product translate well to these other Geos? Are there more enterprise concerns or fewer? Or is it kind of just comparable to the US market? Yeah, I think there's two trends. One is that AI is just such a phenomenon that every buyer out there has tried chat GPT or whatever. And so there is a lot of just top-down pressure from boards and sea sweets to just get moving on something. And if you think about a lot of these businesses, they're like, hmm, how do we adopt AI? Well, it's, you know, it's some coding agents and like let's do customer service because those are like the oddest ones. And so there is a lot of pull to your points. And then the other trend is that the language is a lot easier with AI. So in the past, maybe a blocker would be, oh, my language just doesn't work in, oh, my app just doesn't work in German or picture language, right? But now it's a lot easier to adapt to your app. So I think for those reasons, international has been a lot faster at the same time, right? Like we also want to make sure that we're not getting spread too thin. And so it's kind of this balance where it's like very unclear what the perfect answer is. But you know, we've kind of made a determination of like which markets were okay with really investing in. And if we're going to invest, we're going to really invest. And generally those markets are ones where we already just like naturally picked up some customers out of the US office or something. And then now it's like time to deploy people. Because yeah, even though these there are these trends that make it easier to go internationally, like, you know, there's also other things that you don't really think about, right? Like, you have that data residency. There's all these things where there's like local competitors, which are just know the market a lot better. And so you have to navigate that well. Yeah, absolutely. I think there is a like a space for like local competitors to the large categories that we know about. I mean, there is space for them. I think the question is like long term, is there consolidation? And it's it's you can say that for both the geographies, but also verticals. And then also markets, right? So like there are going to be people that find their niche in different places. Like our view, the reason why we've kind of built a horizontal is that we believe that in our space, the winners are going to be horizontal. There's just not that much that is like super verticalized that where like you could see a pure vertical solution surviving. And historically, that's just been true in our space, right? Like Salesforce, very horizontal. And that's very horizontal. Like all these solutions are very horizontal. Because you just gain more from having that scale and having a very robust and deep product than you do from like having very vertical specific features. And over time, we're also going to build vertical features into it. But our view is that like from our vertical and sort of market point of view, there will be consolidation. Just sort of on related to consolidation, but maybe not just geographic consolidation, I'm curious if AI concierge becomes really the interface between the business and its customers. Do you see like how does CRM and all of these, you know, sort of traditional data, very important systems of record, right, for the end customer? How do those evolve in this new world? And do you see any consolidation there in terms of your own role? Yeah, I don't think it's necessarily an either or, right? Because ultimately what we're doing here is trying to democratize access or availability of these concierge, right? Which is if you were going to a business where you were spending $100,000 a year, you would get the most personalized treatment there and exactly who you are, what your preferences were. They want to help you, you know, shop, for instance, they'll shut the whole store down for you. Actually, I don't know if they do that for a hundred of them. But you get the job. You've never tried that before. However, if you're at a business where you're spending $10, they can't do this for you because they can't make it economical, right? Yeah, it is not a lack of desire to do this for their customers. Right. It is just that the unit economics don't support that. And so effectively, all we're doing here is saying, okay, if you couldn't give them that experience for 10 cents, all of a sudden, it is economical and they'd want to do that. Now, to the point of, does that mean CRMs go away? I mean, my answer is no, because if you have a company today where the concierge is a human being, they still write your info in a CRM so they can track it for later. And I think when we have these agents, they will need somewhere to put that information. So I don't necessarily think those kind of auxiliary pieces of software go away because for us, as we're building these concierge, our goal really is, how do we create that great experience? We will still need places to put that somewhere, right? Yeah. I actually think CRMs could do quite well. They'll be slight different in the sense of, like CRMs are kind of databases in a way. And the frustration people have with them sometimes is that the interfaces are really difficult to use, etc. But in the future, maybe the agents are just using the interfaces and you don't even have graphical interfaces or whatever. And so the CRMs are still very valuable because they hold the source of truth. And then the agents, they're just getting pinged a lot more because the agents are using them. So that is one possible world. So that's kind of bullish on CRMs. And for us personally, like so far, what's sticking on, there's like, we have zero desire to build a CRM because we think there's so much to do in the agentic layer. And that's where we want to focus. Yeah. Okay. So SAS is not dead. Maybe last thing is we were chatting right before we start recording that you guys have done some little AI experimentation on your own as you just figure out how to make your own life more productive and effective. And I'll show you something you've done something kind of interesting. Yeah. You know, I think the models have gotten very smart over the last several years. And you know, the two of us will often use it to brainstorm ideas because they are genuinely very smart at coming up with great ideas. However, the bottleneck, I realize at least for a lot of the work that I do is business context, right? There's a lot of context for every idea around, okay, the constraints that we have, the goals that we're going for and things I've had that is difficult to re-explain to the agents every time. And so I actually spent a while building agents for myself to capture all the business context well. So it just kind of looks over my shoulder all the time and is constantly compiling context on, oh, here are the people that we've hired. Here are the people that we need to hire. Here are the deals that we're working on. Here are the problems. Here are the current challenges that we have. So that later on, I can just go to it and say, hey, there's this new person that we're thinking of hiring. What do you think? And now it's able to automatically be an over. Well, we had two candidates that were very similar. And you know, these candidates had these kind of drawbacks or skills that they didn't have. And so if we hire this person, it's going to be another person that has those kind of exact same things. So we need someone complimentary. So this is probably not the best person to have. Or, you know, in this deal, we're barreling down a very similar path because we didn't validate these things early enough. So this time we should validate them slightly earlier. So it's really all about how do you capture context well? Because then, you know, my job is making decisions with lots of context. So if I can outsource that more and more to a model, maybe I can put myself on a job quicker. Sounds like you created Jesse. There we go. I do think, like one of the big struggles being a solo founder is you don't have anyone's bounce ideas off of. So you just arrive at conclusion with Lussler. And you guys were both some of the others in the past. Right. Yeah. So like us working together, I think the reason it was so much easier is that we could, you know, iterate on things much faster. You can just talk something out. And yeah, now you've talked to Fable or whatever. It's like, it's pretty good. Honestly, it seems like very original. Because in the past, it's just like they just kind of are kind of sick offence, right? Where they just agree with you. That's a good idea. But now they're like, no, that's a bad idea. I actually, for a while, Mark at one point had posted the prof that he used for Claude. It's like, you know, disagree with me. Be very nervous. Things like that. So I actually used that for one. It was great. And the funny story on this was I really enjoyed it because it would agree with me very aggressively. It would disagree with me very aggressively. And I think it's my wife and I was like, Oh, this is great. You should use it. She put it on. And then, you know, a day later, she was like, Oh, Claude was being so mean to me all day. I had to turn it off. I'm like, just kept telling me. Yeah, it was disagreeing on me so aggressively. Amazing. Well, since we're on this topic of founders using a AI to do things. I have to bring up the whole AI Slop Fiasco, maybe is a strong word that Brian Chesky just went through. And I want to bring it up because you both have grown your profile a lot over these last few years. And, you know, I sort of mentioned just you at these pieces that were just hitting the zeitgeist of the discussion that everyone wanted to have and came in with a very differentiated take. And, you know, that's the kind of stuff that we see exactly hit on, you know, you're hitting the conversation right at the, you know, right message, right time, unique message, right time. Do you use AI for writing? And separately, um, this is a more of a meta question, but how important is X and like the sentiment on X to you? Uh, so yeah, for most of our early days, it was mostly linked in with the reasoning of like, Hey, customers are linked in, you know, we're not going to get someone seeing us on Twitter and they're coming as a customer probably. But I think the, the sort of learning we had with X is, or at least I was kind of reflecting on it. X is kind of like the timeline that people talk about. And most of that timeline is not really about your company. Like, if you're, if you're very like just like promoting yourself on X, like it's going to get no traction whatsoever, totally. But it is sort of like a single timeline that everyone reads. So I kind of like, it kind of like, my controls everyone to be thinking about the same thing. Yes. And that's where it's valuable to have some say in it. And, uh, who was I listening to? There's like, there's a guy like Jeremy Giffan or something who's on Patrick O'Shaussi's podcast. And he was like doing, he like made some claim about how like, you know, it used to be that like the, the, the big status symbols in the world is like, everyone wants to be a billionaire because like, he calls it like the priest class or whatever. So it's like, you know, billionaires are the priest class. But nowadays, actually, it's when people become billionaires, now they want to become X influencers. Because like, those people hold the real power because they can like influence what the whole world's thinking about. So that's like one reason to have some presence on X, I suppose. And when we, when now when I post on X, I don't really post about like decky on specifically, it's more or more about, you know, our thoughts on what is happening. So I think that's, that's like a good, good way to do it. And it's very different. You write it all yourself, right? Yeah. Um, I mean, AI is good for, um, sort of helping you brainstorm like what topics to write about. I think that's, but that's pretty good. But I think that was, that's kind of like a learning that like LinkedIn and X work very differently. You can just like come up with something cool and like post the same thing on both. Cause like very few things, like do well on both. Yeah. So like, LinkedIn's really good for, you know, classic stuff. You're making announcements and talking about your product and, you know, your fundraise or whatever. I guess you can do fundraise on X as well. But X is a lot more about kind of like place yourself on top of that single timeline that everyone's on. And is that, so is it too much of a simple, you know, simplification to say X for hiring, especially, you know, AI research, talent, et cetera, um, maybe ecosystem as well. And then LinkedIn more for enterprise customers. Or are you actually seeing enterprise CIOs pay attention to X heart attributions hard, obviously. Yeah, it's hard. Cause, you know, I think you could say like, Oh, well, you know, I, I made a post on X and then, you know, the people on the all-in podcast were talking about it. Like CIOs, definitely, yes. So it's like a, like indirectly unsure. We got like some eyeballs from CIOs is the CIO themselves like scrolling X all day, maybe not. But if you are part of that major time line, then you have, there's all of these like secondary effects. And then, and then like reporters will reach out to like from my mainstream media. And then if they write about you, then like, Austin was on the New York Times recently. You know, it's like if they write about you, then those for sure get, I also, yeah, you were, um, but it's open source stuff. I don't can't release only on X. So, yeah. You didn't put the next one. I didn't see it. I guess maybe the last question, we just, we touched on like hot button X topics. Um, one, and this is kind of a serious one actually, but it sort of re-entered the narrative, I think with, um, you know, the anthropic video and it's just sort of maybe never left the narrative. But it's on this concept as progress gets better around jobs. Um, and the messaging around that, um, it's a sensitive topic, obviously, but it's interesting because I really think customer support was maybe the first end-to-end use case where you could really take an entire job. Um, I'm sorry, do an entire job, I should say, take as the wrong word versus coding was always, you know, pair programming to start with. Um, how has that, you know, we sort of joked before, but I, you know, we weren't really joking that, um, AI is actually creating jobs. I'm curious, how do you turn that narrative on its head that folks are just losing their jobs, right? Like, do you see the up-leveling of folks with AI where, you know, maybe they were doing this job and now they're doing something else? Yeah. I mean, we, we see this all the time. Uh, because, uh, if you were called earlier on, when we were talking about, hey, what are we truly doing? Um, we found that in a, for a lot of our customers, there's actually just more demand for things like customer support than their supply, right? Where companies realize that they're like, okay, if our cost of doing customer support drops by 30%, most of them are not just immediately saying, okay, now what I will do is, you know, let go of 60% of my team. They're saying, okay, now that this thing, which is clearly valuable for my customers is much cheaper, let me do more of it. So that my customers will retain for longer so that, you know, they don't show it off as much so that they activate sooner things like that. You know, we had a customer in the early days, um, they, and this was, you know, probably two and a half years ago at this point, um, where they said, you know, our ticket volume, you know, the amount of customer support inquiries that we get, uh, per month was I think, I think there's like 50,000 a month or something based on, uh, you know, the existing surfaces that they had. Once they started using us, they said, wow, turn that our customers have a lot of problems. They said, let us make support more easily accessible, right? So instead of just being in one part, like buried within a support panel, they're like, let's put support on every page and let's make it more prominent in places where people are more likely to get stuck. Let's, uh, immediate support for even free users rather than paying users, right? So because of this kind of, there's more kind of related demand for support. Yeah. Uh, then there are supply. So automating things doesn't necessarily result in just kind of people laying off their entire teams. That may be the best example of Jevin's paradox in real life that I've heard. So, yeah, I think it's like AI will, uh, kill jobs, but not careers in a way because like those jobs that are being done currently should not be done by humans. Like they're very mundane and menial. It's like, it's like a super high volume use case and people are just like, kind of thinking out the phone is like, okay, let me click here, click here and like, okay, here's an answer, right? And that should be done by AI. But there is actually a like a near infinite amount of things that people could be doing to make their customers happier and, you know, take care of them more. And so people end up doing those things. And more and more of the mundane repeatable things will get each and up by AI. So that, that's, that's what we think will happen. Just I think on a podcast, maybe Patrick O'Shaughnessy's podcast a couple months ago, you had mentioned that even your customers, when they've been using BPO's for customers, support and said, you haven't actually seen like layoffs at the BPO just turns out that those employees have gone and done other things instead. Is that still true? Oh, I mean, it's, it's, it really depends on the situation. So there are definitely scenarios where people use their BPO's a lot less or don't need the BPO anymore. The other situations where they are not in cost cutting mode whatsoever. And their goal is to either their business is growing so quickly that they don't want to scale their operations along with their growth. And so with that, you go on, they can kind of like keep the flats or, or not or it's actually we still need people, but now there's all these other things they could be doing. And it could be more revenue generating things. You know, it's, that's like a big area for, for us Siemens, like as the AI matures, we first start with these costs cutting use cases because those are easy, but then like revenue generating use cases should also be able to be done through this conversational interface. So yeah, there's, it really depends on the customer, but you know, we definitely have customers that made massive changes. Actually, I like to end it on that uplifting note. And just to repeat, what Jesse said, it may kill jobs, but not careers. I love that. Thank you so much for joining us, guys. A pleasure to have you. Thanks for listening to this episode of the A60Z podcast. If you like this episode, be sure to like, comment, subscribe, leave us a rating or review, and share it with your friends and family. For more episodes, go to YouTube, Apple Podcasts, and Spotify. Follow us on X at A16Z, and subscribe to our substack at a16z.substack.com. Thanks again for listening, and I'll see you in the next episode. As a reminder, the content here is for informational purposes. is only, should not be taken as legal business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A-16Z fund. Please note that A-16Z and its affiliates may also maintain investments in the company's discussed in this podcast. For more details, including a link to our investments, please see A-16Z.com/disclosures. [Music]

Podcast Summary

Key Points:

  1. Decagon shifted most of its AI workflow to open-source models to reduce latency and improve performance on specific tasks, despite initial reliance on frontier models.
  2. Smaller, fine-tuned open-source models outperform large, general-purpose models on narrow tasks, delivering better accuracy, speed, and cost efficiency.
  3. The company built a dedicated research team (Decagon Labs) to tune and optimize open-source models, emphasizing custom evaluation frameworks tied to real business outcomes.
  4. Enterprise AI progress is driven not just by model intelligence, but by product design and the ability to manage complex workflows, business logic, and compliance.
  5. Decagon’s core moat lies in its deep integration with enterprise operations—enabling secure, scalable, and auditable AI deployment across legacy systems and processes.
  6. The future of enterprise AI involves a hybrid approach
  7. Forward-deployed engineers are evolving from one-off consultants into product contributors who drive long-term productization of AI workflows.
  8. Decagon’s success stems from a product-led, sales-driven model that enables rapid iteration, transparency, and control—key for large enterprises hesitant to adopt AI.

Summary:

Decagon transitioned its AI stack to open-source models to achieve faster, more cost-efficient performance on specific business tasks, such as customer support and sales qualification, where smaller, fine-tuned models outperform large general-purpose models. Despite early use of frontier models, the company now relies on open-source models for 90% of workflows due to superior latency and task-specific accuracy. This shift was enabled by a dedicated research team that builds custom evaluation frameworks to measure real-world business outcomes, not just model performance.

Decagon’s long-term advantage lies in its deep integration with enterprise operations—managing compliance, legacy systems, and business logic—making AI deployment secure, scalable, and controllable. The company’s product-led, sales-driven approach allows rapid iteration and transparency, which resonates with large enterprises that fear AI black boxes. As AI evolves, Decagon maintains its relevance by focusing on operational workflows and business process automation, not just customer service.

While frontier models remain valuable for exploratory tasks, the company’s model is built to be flexible, scalable, and deeply embedded in real-world enterprise processes. Ultimately, Decagon’s future success hinges on its ability to productize AI workflows and serve as a trusted, transparent partner in enterprise digital transformation.

FAQs

Decagon uses open-source models primarily for lower latency and better performance on specific tasks. Smaller, specialized models can outperform larger models in targeted workflows while being faster and cheaper, especially for real-time voice agents.

Today, 90% of Decagon’s workflows run on open-source models for speed and cost efficiency, while the remaining 10% use closed-source models for new, experimental products or tasks requiring broad general intelligence and creative problem-solving.

Decagon Labs acts as a model factory, continuously fine-tuning and evaluating open-source models to create custom agents tailored to specific business processes. It enables rapid iteration and adaptation to evolving model capabilities.

Decagon builds robust systems that include business logic, customer context, security checks, compliance, and real-time monitoring. These ensure agents operate safely, accurately, and in alignment with enterprise needs, even as models evolve.

Decagon’s agents are not just wrappers—they are designed to follow complex business processes. They include agent operating procedures, tools, testing, and self-improvement capabilities, enabling them to learn, adapt, and optimize over time.

Decagon developed an internal agent (called 'Duet Autopilot') that can analyze conversations, identify pain points, and automatically generate improvements, reducing the need for manual AOP writing and iterative testing.

Chat with AI

Loading...

Pro features

Go deeper with this episode

Unlock creator-grade tools that turn any transcript into show notes and subtitle files.